The algorithmic tendency

Artificial intelligence is arguably the technological advancement of our time. The widespread ramifications are yet unknown, and commentators differ in their predictions of the future. Technology is an unfolding experiment, as it always has been.

In this article, artificial intelligence refers primarily to generative artificial intelligence, particularly large language models capable of generating, synthesising and interpreting language in response to natural-language prompts. My concern is not with artificial intelligence in its entirety, but with the increasing use of such systems to undertake cognitive tasks previously performed by clinicians.

Although technological advancement has been the driving force behind our developments as a civilisation, we must endeavour to maintain some discretion, to maintain our critical eye. Although the course we are taking often feels out of our control, we make the decisions – and either enjoy or suffer the consequences – every step of the way. But how do we make wise decisions when we do not feel in control? There is a sense of letting the experiment unfold in whichever way leads to more profit and titillation. But to what end? Or is that the end? The image of a runaway train sometimes comes to mind.

As a psychiatrist, I am concerned about the direction we are moving towards. I am also concerned about where we find ourselves at present. Clinicians are already using artificial intelligence tools to draft clinic letters, generate formulations and summarise patient histories. The modern clinician may soon leave the patient alone in the clinic room – metaphorically, perhaps eventually literally. Nonetheless, as psychiatrists we are required to use rational, empirical and intuitive faculties. We rely on a knowledge base built through concentrated study, and the development of intuitive instincts from within the clinical encounter.

Artificial intelligence did not create algorithmic psychiatry. Protocols, diagnostic criteria, risk tools and increasingly prescriptive clinical guidelines have long created tensions between standardisation and individual clinical judgement. Nor did artificial intelligence inaugurate the fragmentation of attention: smartphones, digital media and screen-based reading preceded generative artificial intelligence. Artificial intelligence arrives at a moment when psychiatry is already increasingly susceptible to algorithmic practice, and may accelerate that tendency to an unprecedented degree.

There are many angles from which to approach these issues, and many aspects that deserve thorough examination and thoughtful critique. For this article, I shall offer some observational reflections and questions around the unfolding effects of artificial intelligence on the field of psychiatry: on clinicians, on their patients and on the clinical encounter in which the two meet. What effect might artificial intelligence have on our ability to think, on our ability to empathise and connect with the patient, and to construct clinical formulations based on our understanding of the patient, our intuition and scientific knowledge?

These questions could, of course, have been posed at various points during recent technological innovation. Consider, for example, the screen: the small, two-dimensional surface upon which we now depend, offering an apparently limitless field of content and communication. A bounded physical object has become a portal to seemingly infinite possibilities for our attention; there is always somewhere else to go. Such technologies have already altered our relationship with attention, communication and the physical world. The hitherto uncharacterised effects of artificial intelligence must therefore be understood within this broader technological context. Yet there is reason to ask whether something qualitatively different is now taking place. The answers will become clearer over time as both experiential and empirical data emerge, but I believe we are already at a point when existing experience and observations can inform a critique.

Cognitive outsourcing

First, I will give the most immediate account available to me of these technologies. From my personal experience, the effects are somewhat insidious but now feel profound and inescapable. My enjoyment of reading quietly, my capacity for concentrated learning, for engrossment in a journey of discovery, for a stillness of mind – all have diminished or been extinguished. Indeed, the writing of this short article is a struggle in a way that feels almost nauseating, with a constant effort to resist the compulsion to redirect my attention elsewhere, to pick up my phone, to ask artificial intelligence a question, to skip to a conclusion.

The causal mechanisms are undoubtedly multifactorial and elusive. One could argue that this is a simple case of unmasked attention-deficit hyperactivity disorder (ADHD). However, there is the foundational principle that the individual cannot be separated from their environment. And it is the environment that has changed so rapidly and fundamentally, interacting with us in ways that feel unfamiliar. Anecdotally, I have found this pattern of psychological effects, to a greater or lesser extent, to be widely felt among colleagues.

Recent experimental work has introduced the term ‘cognitive debt’ to describe differences in cognitive engagement associated with artificial intelligence-assisted task completion. ^{Reference Kosmyna, Hauptmann, Yuan, Situ, Liao and Beresnitzky1} Although preliminary and insufficient to establish persistent cognitive impairment, such findings raise a question of particular relevance to psychiatry: what capacities might be diminished when cognitive work is repeatedly outsourced?

Recent longitudinal research distinguishes between ‘dependent’ cognitive offloading, in which core thinking (‘cognitive agency’) is delegated to generative artificial intelligence, and ‘autonomous’ offloading, where artificial intelligence supports rather than substitutes the individual’s cognitive effort. ^{Reference Zhu, Li, Dong, Chang and Fan2} Dependent offloading was associated with poorer perceived cognitive functioning, although the findings remain correlational rather than evidence of cognitive decline.

The stratification of any such effects also requires examination. Are certain groups, including those with ADHD or other forms of neurodiversity, more or less sensitive to them? Is there a generational sensitivity? Might a population whose cognitive foundations were built during the ‘analogue age’ experience these technologies differently from a younger generation coming of age within the artificial intelligence era?

The intoxication of immediacy

What, then, is different about artificial intelligence? The metaphor of technology as a tool no longer applies. Artificial intelligence is generative: it imitates executive function and offers solutions de novo. As we integrate artificial intelligence into our profession, we become executors of algorithms, engulfed in information. The patient is lost, fragmented into data points, feeding a non-human intelligence that processes the data and offers information in exchange, instantly. Time collapses in on itself. Algorithmic decisions are instant, so thought is obsolete. Comprehension takes time. But when the medical system aspires to efficiency and outcomes, there is no time for understanding. The practice of psychiatry is said to involve a leap to bridge that seemingly incomprehensible void between the physician and the patient. ^{Reference Jaspers3} But the leap takes time.

The usefulness of artificial intelligence is not in question. It may reduce administrative burden, synthesise lengthy histories and support aspects of clinical reasoning. Indeed, its utility is precisely what makes the question urgent. Technologies that are useless require no resistance. The danger lies in the ease with which assistance can become substitution: cognitive work once undertaken by the clinician can disappear without the clinician noticing what has been lost.

Dwelling in the unknown

Psychoanalysis can perhaps serve here as a motif: a clinical tradition that deliberately preserves time, uncertainty, interpersonal attention and non-algorithmic understanding. These qualities are not unique to psychoanalysis and belong also to the wider humanistic tradition within medicine. Yet psychoanalysis offers a particularly distilled expression of them. There is a constitutive frame within which psychoanalysis must provide a private meeting of two human beings; in essence, with few exceptions, that is all it is. From that point on, comprehension (communion) can take place.

So now, more than ever, we could re-examine psychoanalysis, alongside other schools of humanistic practice, as a resistance against the surrender and outsourcing of our intelligence and human faculties.

One often-quoted psychoanalytic principle, extending to psychiatry as a whole, is the ability to tolerate ambivalence – to dwell in the unknown. Artificial intelligence does not tolerate such states and offers no such dwelling. Information crowds ambivalence into oblivion. There is information that frees us and information that traps us; there is a risk that we will venture deep into the territory of the latter. To comprehend the Other, one must tolerate time, waiting and thereby accept the Otherness of the patient through the passage of time.

In contrast to the psychoanalytic paradigm, artificial intelligence could be said to both engulf absence (filling the unknown with information) and collapse temporality (offering immediacy), thereby changing our relationship with our reality (space and time), our place in the world and, by implication, what it means to be human. The stakes are high.

The shifting norm

Normality is both a problematic and essential vocabulary for psychiatrists. We are constantly referencing the proverbial ‘norm’, whether explicitly or not. Yet normality resists normalisation. It bends and kinks with the times. Psychiatry must therefore attend not only to pathology, but also to the changing conditions of experience from which the normal itself emerges. As psychiatrists, we encounter this shifting normality within both ourselves and our patients.

Laing’s ‘ontological insecurity’, conceived as an existential account of psychosis, ^{Reference Laing4} acquires a different resonance in a world of fragmented attention, compulsive communication and increasingly unstable relations with space, time and the physical world. What was once described at the margins of human experience may, in attenuated form, move towards the norm. Artificial intelligence did not create this condition, but arrives within – and may intensify – a technological environment already destabilising the Self.

There is a risk that, rather than psychiatry integrating insights from psychoanalysis and the wider humanistic tradition, there is yet more divergence as medical psychiatry becomes increasingly algorithmic and artificial intelligence-driven, while psychoanalysis fades further into esoteric territory, lost in time metaphorically and in practice.

A fleeting vantage point

We exist within a poignant window of time – close enough to be profoundly affected by artificial intelligence but distant enough to feel the change, to detect the deficits and question what we see on the horizon. We are intoxicated enough to feel lucidity fading, but just sufficiently lucid to know that we are intoxicated. This may not last.